Asking AI for legal advice could seal your defeat in a legal proceeding

Jason Clark • June 9, 2026

In a lawsuit or criminal prosecution, questions posed to an artificial intelligence assistant, and the AI assistant’s replies, might be used against you.

Jason Clark

By Jason Clark


As with many forms of emerging technology, the uses for artificial intelligence (AI) are outpacing its practical and legal consequences. Using AI as a legal research tool or sounding board is no exception.


Early this year, in U.S. v. Heppner, a U.S. District Court judge in New York ruled that written exchanges between a criminal defendant and a generative AI platform were not protected by attorney-client privilege or the work product doctrine. (Attorney-client privilege protects confidential communications between a client and their attorney. The work product doctrine protects an attorney’s trial preparation materials – such as notes, legal strategies, research, etc. – from being discovered by opposing counsel.)

Because Heppner is a “case of first impression,” the court’s ruling is by no means carved in granite. For now, however, it flashes a sign of caution for members of the public – and their attorneys – in using AI to perform legal research in adversarial matters.


Background. The case involved the criminal fraud prosecution of Bradley Heppner. To assess his legal situation and possible defenses, Mr. Heppner directly posed questions to the AI platform Claude. He then shared with his attorneys the written results of his exchanges with Claude.


In the subsequent search of Mr. Heppner’s home, FBI agents seized his computer, which contained his AI-generated documents. Mr. Heppner’s attorneys moved to shield those documents from prosecutors, arguing that the documents were privileged communications and protected work product.


The judge in the case rejected those arguments and granted the government’s motion to access Mr. Heppner’s AI-generated documents. The judge cited several key reasons, including:


  • AI is not an attorney. As a consequence, Mr. Heppner’s exchanges with Claude about his legal issues are not protected by attorney-client privilege.
  • He forfeited any privilege. Even if the information Mr. Heppner shared with Claude was originally confidential, he waived any attorney-client privilege when he disclosed the issues with the third-party AI platform.
  • He had no expectation of confidentiality. Mr. Heppner used a public version of the AI assistant, and Claude’s privacy policy allowed it to collect user inputs for model training and to share data with third parties.
  • He should have let his attorneys do his legal work. Mr. Heppner’s AI documents were prepared on his own volition, not by or at the request of an attorney in anticipation of litigation.


Legal Implications. The Heppner ruling delivers a clear warning to attorneys and their clients. It affirms that inputting case details, trial strategy, or confidential client communications into publicly accessible generative AI tools (such as Claude, ChatGPT, or Gemini) can waive privilege and expose the resulting communication to the other side in the course of litigation. 


For contractors and other business owners, this ruling serves as an important reminder that attorney-client communications and attorney work-product materials should always be maintained in the strictest confidence; sharing such items with third parties – including AI platforms or internet-based search engines – can destroy those protections.